用 Opus 5.5 当导演剪辑的 Lightspeed 宣传片
作者用 Opus 5.5 在 Cowork 里写分镜和剪辑代码，配 HyperFrames、Kling、ElevenLabs 做出一支宣传片，并反思 AI 视频救不了内容策略。
Opus 5.5 generated videos won't save your content strategy. If anything they'll make it worse. You'll hit local maxima but feel you've cracked Coke's secret sauce. Gen AI tools work best for people who are already good at the task. I write a lot with AI and my writing has gotten better and faster. But I was good at writing long-form pre-ChatGPT. However, I suck at video, and AI doesn't change that. Case in point - this video, which I was making for a Lightspeed project I was working on. When it reached its current stage I felt victorious and thought it was amazing. More amazing was that I did it in one sitting, 10 pm to 3 am, missteps and experiments included. I can share it here because I am no longer using it. Thankfully better sense prevailed and I'm now working with a specialist. The stack, for those interested: > Claude Opus 5.5 in Cowork as director and editor - shot list, prompts, and the whole edit written as code > HyperFrames for the timeline and motion graphics (HTML + GSAP rendered to video) > Kling VIDEO 3.0 and IMAGE 3.0 for footage, Gemini for reference stills > ElevenLabs for the radio voices > Envato for stock footage and sound effects > ffmpeg + OpenCV for the sound mix, colour grade, and tracking the radio needle into the marker Don't get me wrong. I learned so much more about using AI to make videos in those 5 hours than I would have learned by reading about it for 50 hours. But that's where it stops. Don't be misled by the X echo chambers. Sometimes, the best time to grow rice is when everyone else is growing wheat. So maybe when everyone is making an Opus 5.5 video, it is the best time to write long-forms.